net/core/sysctl_net_core.c

   1 /* -*- linux-c -*-
   2  * sysctl_net_core.c: sysctl interface to net core subsystem.
   3  *
   4  * Begun April 1, 1996, Mike Shaver.
   5  * Added /proc/sys/net/core directory entry (empty =) ). [MS]
   6  */
   7
   8 #include <linux/mm.h>
   9 #include <linux/sysctl.h>
  10 #include <linux/module.h>
  11 #include <linux/socket.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/ratelimit.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/init.h>
  16 #include <linux/slab.h>
  17 #include <linux/kmemleak.h>
  18
  19 #include <net/ip.h>
  20 #include <net/sock.h>
  21 #include <net/net_ratelimit.h>
  22 #include <net/busy_poll.h>
  23 #include <net/pkt_sched.h>
  24
  25 static int zero = 0;
  26 static int one = 1;
  27 static int ushort_max = USHRT_MAX;
  28
  29 static int net_msg_warn;        /* Unused, but still a sysctl */
  30
  31 #ifdef CONFIG_RPS
  32 static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
  33                                 void __user *buffer, size_t *lenp, loff_t *ppos)
  34 {
  35         unsigned int orig_size, size;
  36         int ret, i;
  37         struct ctl_table tmp = {
  38                 .data = &size,
  39                 .maxlen = sizeof(size),
  40                 .mode = table->mode
  41         };
  42         struct rps_sock_flow_table *orig_sock_table, *sock_table;
  43         static DEFINE_MUTEX(sock_flow_mutex);
  44
  45         mutex_lock(&sock_flow_mutex);
  46
  47         orig_sock_table = rcu_dereference_protected(rps_sock_flow_table,
  48                                         lockdep_is_held(&sock_flow_mutex));
  49         size = orig_size = orig_sock_table ? orig_sock_table->mask + 1 : 0;
  50
  51         ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
  52
  53         if (write) {
  54                 if (size) {
  55                         if (size > 1<<30) {
  56                                 /* Enforce limit to prevent overflow */
  57                                 mutex_unlock(&sock_flow_mutex);
  58                                 return -EINVAL;
  59                         }
  60                         size = roundup_pow_of_two(size);
  61                         if (size != orig_size) {
  62                                 sock_table =
  63                                     vmalloc(RPS_SOCK_FLOW_TABLE_SIZE(size));
  64                                 if (!sock_table) {
  65                                         mutex_unlock(&sock_flow_mutex);
  66                                         return -ENOMEM;
  67                                 }
  68
  69                                 sock_table->mask = size - 1;
  70                         } else
  71                                 sock_table = orig_sock_table;
  72
  73                         for (i = 0; i < size; i++)
  74                                 sock_table->ents[i] = RPS_NO_CPU;
  75                 } else
  76                         sock_table = NULL;
  77
  78                 if (sock_table != orig_sock_table) {
  79                         rcu_assign_pointer(rps_sock_flow_table, sock_table);
  80                         if (sock_table)
  81                                 static_key_slow_inc(&rps_needed);
  82                         if (orig_sock_table) {
  83                                 static_key_slow_dec(&rps_needed);
  84                                 synchronize_rcu();
  85                                 vfree(orig_sock_table);
  86                         }
  87                 }
  88         }
  89
  90         mutex_unlock(&sock_flow_mutex);
  91
  92         return ret;
  93 }
  94 #endif /* CONFIG_RPS */
  95
  96 #ifdef CONFIG_NET_FLOW_LIMIT
  97 static DEFINE_MUTEX(flow_limit_update_mutex);
  98
  99 static int flow_limit_cpu_sysctl(struct ctl_table *table, int write,
 100                                  void __user *buffer, size_t *lenp,
 101                                  loff_t *ppos)
 102 {
 103         struct sd_flow_limit *cur;
 104         struct softnet_data *sd;
 105         cpumask_var_t mask;
 106         int i, len, ret = 0;
 107
 108         if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 109                 return -ENOMEM;
 110
 111         if (write) {
 112                 ret = cpumask_parse_user(buffer, *lenp, mask);
 113                 if (ret)
 114                         goto done;
 115
 116                 mutex_lock(&flow_limit_update_mutex);
 117                 len = sizeof(*cur) + netdev_flow_limit_table_len;
 118                 for_each_possible_cpu(i) {
 119                         sd = &per_cpu(softnet_data, i);
 120                         cur = rcu_dereference_protected(sd->flow_limit,
 121                                      lockdep_is_held(&flow_limit_update_mutex));
 122                         if (cur && !cpumask_test_cpu(i, mask)) {
 123                                 RCU_INIT_POINTER(sd->flow_limit, NULL);
 124                                 synchronize_rcu();
 125                                 kfree(cur);
 126                         } else if (!cur && cpumask_test_cpu(i, mask)) {
 127                                 cur = kzalloc_node(len, GFP_KERNEL,
 128                                                    cpu_to_node(i));
 129                                 if (!cur) {
 130                                         /* not unwinding previous changes */
 131                                         ret = -ENOMEM;
 132                                         goto write_unlock;
 133                                 }
 134                                 cur->num_buckets = netdev_flow_limit_table_len;
 135                                 rcu_assign_pointer(sd->flow_limit, cur);
 136                         }
 137                 }
 138 write_unlock:
 139                 mutex_unlock(&flow_limit_update_mutex);
 140         } else {
 141                 char kbuf[128];
 142
 143                 if (*ppos || !*lenp) {
 144                         *lenp = 0;
 145                         goto done;
 146                 }
 147
 148                 cpumask_clear(mask);
 149                 rcu_read_lock();
 150                 for_each_possible_cpu(i) {
 151                         sd = &per_cpu(softnet_data, i);
 152                         if (rcu_dereference(sd->flow_limit))
 153                                 cpumask_set_cpu(i, mask);
 154                 }
 155                 rcu_read_unlock();
 156
 157                 len = min(sizeof(kbuf) - 1, *lenp);
 158                 len = cpumask_scnprintf(kbuf, len, mask);
 159                 if (!len) {
 160                         *lenp = 0;
 161                         goto done;
 162                 }
 163                 if (len < *lenp)
 164                         kbuf[len++] = '\n';
 165                 if (copy_to_user(buffer, kbuf, len)) {
 166                         ret = -EFAULT;
 167                         goto done;
 168                 }
 169                 *lenp = len;
 170                 *ppos += len;
 171         }
 172
 173 done:
 174         free_cpumask_var(mask);
 175         return ret;
 176 }
 177
 178 static int flow_limit_table_len_sysctl(struct ctl_table *table, int write,
 179                                        void __user *buffer, size_t *lenp,
 180                                        loff_t *ppos)
 181 {
 182         unsigned int old, *ptr;
 183         int ret;
 184
 185         mutex_lock(&flow_limit_update_mutex);
 186
 187         ptr = table->data;
 188         old = *ptr;
 189         ret = proc_dointvec(table, write, buffer, lenp, ppos);
 190         if (!ret && write && !is_power_of_2(*ptr)) {
 191                 *ptr = old;
 192                 ret = -EINVAL;
 193         }
 194
 195         mutex_unlock(&flow_limit_update_mutex);
 196         return ret;
 197 }
 198 #endif /* CONFIG_NET_FLOW_LIMIT */
 199
 200 #ifdef CONFIG_NET_SCHED
 201 static int set_default_qdisc(struct ctl_table *table, int write,
 202                              void __user *buffer, size_t *lenp, loff_t *ppos)
 203 {
 204         char id[IFNAMSIZ];
 205         struct ctl_table tbl = {
 206                 .data = id,
 207                 .maxlen = IFNAMSIZ,
 208         };
 209         int ret;
 210
 211         qdisc_get_default(id, IFNAMSIZ);
 212
 213         ret = proc_dostring(&tbl, write, buffer, lenp, ppos);
 214         if (write && ret == 0)
 215                 ret = qdisc_set_default(id);
 216         return ret;
 217 }
 218 #endif
 219
 220 static int proc_do_rss_key(struct ctl_table *table, int write,
 221                            void __user *buffer, size_t *lenp, loff_t *ppos)
 222 {
 223         struct ctl_table fake_table;
 224         char buf[NETDEV_RSS_KEY_LEN * 3];
 225
 226         snprintf(buf, sizeof(buf), "%*phC", NETDEV_RSS_KEY_LEN, netdev_rss_key);
 227         fake_table.data = buf;
 228         fake_table.maxlen = sizeof(buf);
 229         return proc_dostring(&fake_table, write, buffer, lenp, ppos);
 230 }
 231
 232 static struct ctl_table net_core_table[] = {
 233 #ifdef CONFIG_NET
 234         {
 235                 .procname       = "wmem_max",
 236                 .data           = &sysctl_wmem_max,
 237                 .maxlen         = sizeof(int),
 238                 .mode           = 0644,
 239                 .proc_handler   = proc_dointvec_minmax,
 240                 .extra1         = &one,
 241         },
 242         {
 243                 .procname       = "rmem_max",
 244                 .data           = &sysctl_rmem_max,
 245                 .maxlen         = sizeof(int),
 246                 .mode           = 0644,
 247                 .proc_handler   = proc_dointvec_minmax,
 248                 .extra1         = &one,
 249         },
 250         {
 251                 .procname       = "wmem_default",
 252                 .data           = &sysctl_wmem_default,
 253                 .maxlen         = sizeof(int),
 254                 .mode           = 0644,
 255                 .proc_handler   = proc_dointvec_minmax,
 256                 .extra1         = &one,
 257         },
 258         {
 259                 .procname       = "rmem_default",
 260                 .data           = &sysctl_rmem_default,
 261                 .maxlen         = sizeof(int),
 262                 .mode           = 0644,
 263                 .proc_handler   = proc_dointvec_minmax,
 264                 .extra1         = &one,
 265         },
 266         {
 267                 .procname       = "dev_weight",
 268                 .data           = &weight_p,
 269                 .maxlen         = sizeof(int),
 270                 .mode           = 0644,
 271                 .proc_handler   = proc_dointvec
 272         },
 273         {
 274                 .procname       = "netdev_max_backlog",
 275                 .data           = &netdev_max_backlog,
 276                 .maxlen         = sizeof(int),
 277                 .mode           = 0644,
 278                 .proc_handler   = proc_dointvec
 279         },
 280         {
 281                 .procname       = "netdev_rss_key",
 282                 .data           = &netdev_rss_key,
 283                 .maxlen         = sizeof(int),
 284                 .mode           = 0444,
 285                 .proc_handler   = proc_do_rss_key,
 286         },
 287 #ifdef CONFIG_BPF_JIT
 288         {
 289                 .procname       = "bpf_jit_enable",
 290                 .data           = &bpf_jit_enable,
 291                 .maxlen         = sizeof(int),
 292                 .mode           = 0644,
 293                 .proc_handler   = proc_dointvec
 294         },
 295 #endif
 296         {
 297                 .procname       = "netdev_tstamp_prequeue",
 298                 .data           = &netdev_tstamp_prequeue,
 299                 .maxlen         = sizeof(int),
 300                 .mode           = 0644,
 301                 .proc_handler   = proc_dointvec
 302         },
 303         {
 304                 .procname       = "message_cost",
 305                 .data           = &net_ratelimit_state.interval,
 306                 .maxlen         = sizeof(int),
 307                 .mode           = 0644,
 308                 .proc_handler   = proc_dointvec_jiffies,
 309         },
 310         {
 311                 .procname       = "message_burst",
 312                 .data           = &net_ratelimit_state.burst,
 313                 .maxlen         = sizeof(int),
 314                 .mode           = 0644,
 315                 .proc_handler   = proc_dointvec,
 316         },
 317         {
 318                 .procname       = "optmem_max",
 319                 .data           = &sysctl_optmem_max,
 320                 .maxlen         = sizeof(int),
 321                 .mode           = 0644,
 322                 .proc_handler   = proc_dointvec
 323         },
 324 #ifdef CONFIG_RPS
 325         {
 326                 .procname       = "rps_sock_flow_entries",
 327                 .maxlen         = sizeof(int),
 328                 .mode           = 0644,
 329                 .proc_handler   = rps_sock_flow_sysctl
 330         },
 331 #endif
 332 #ifdef CONFIG_NET_FLOW_LIMIT
 333         {
 334                 .procname       = "flow_limit_cpu_bitmap",
 335                 .mode           = 0644,
 336                 .proc_handler   = flow_limit_cpu_sysctl
 337         },
 338         {
 339                 .procname       = "flow_limit_table_len",
 340                 .data           = &netdev_flow_limit_table_len,
 341                 .maxlen         = sizeof(int),
 342                 .mode           = 0644,
 343                 .proc_handler   = flow_limit_table_len_sysctl
 344         },
 345 #endif /* CONFIG_NET_FLOW_LIMIT */
 346 #ifdef CONFIG_NET_RX_BUSY_POLL
 347         {
 348                 .procname       = "busy_poll",
 349                 .data           = &sysctl_net_busy_poll,
 350                 .maxlen         = sizeof(unsigned int),
 351                 .mode           = 0644,
 352                 .proc_handler   = proc_dointvec
 353         },
 354         {
 355                 .procname       = "busy_read",
 356                 .data           = &sysctl_net_busy_read,
 357                 .maxlen         = sizeof(unsigned int),
 358                 .mode           = 0644,
 359                 .proc_handler   = proc_dointvec
 360         },
 361 #endif
 362 #ifdef CONFIG_NET_SCHED
 363         {
 364                 .procname       = "default_qdisc",
 365                 .mode           = 0644,
 366                 .maxlen         = IFNAMSIZ,
 367                 .proc_handler   = set_default_qdisc
 368         },
 369 #endif
 370 #endif /* CONFIG_NET */
 371         {
 372                 .procname       = "netdev_budget",
 373                 .data           = &netdev_budget,
 374                 .maxlen         = sizeof(int),
 375                 .mode           = 0644,
 376                 .proc_handler   = proc_dointvec
 377         },
 378         {
 379                 .procname       = "warnings",
 380                 .data           = &net_msg_warn,
 381                 .maxlen         = sizeof(int),
 382                 .mode           = 0644,
 383                 .proc_handler   = proc_dointvec
 384         },
 385         { }
 386 };
 387
 388 static struct ctl_table netns_core_table[] = {
 389         {
 390                 .procname       = "somaxconn",
 391                 .data           = &init_net.core.sysctl_somaxconn,
 392                 .maxlen         = sizeof(int),
 393                 .mode           = 0644,
 394                 .extra1         = &zero,
 395                 .extra2         = &ushort_max,
 396                 .proc_handler   = proc_dointvec_minmax
 397         },
 398         { }
 399 };
 400
 401 static __net_init int sysctl_core_net_init(struct net *net)
 402 {
 403         struct ctl_table *tbl;
 404
 405         net->core.sysctl_somaxconn = SOMAXCONN;
 406
 407         tbl = netns_core_table;
 408         if (!net_eq(net, &init_net)) {
 409                 tbl = kmemdup(tbl, sizeof(netns_core_table), GFP_KERNEL);
 410                 if (tbl == NULL)
 411                         goto err_dup;
 412
 413                 tbl[0].data = &net->core.sysctl_somaxconn;
 414
 415                 /* Don't export any sysctls to unprivileged users */
 416                 if (net->user_ns != &init_user_ns) {
 417                         tbl[0].procname = NULL;
 418                 }
 419         }
 420
 421         net->core.sysctl_hdr = register_net_sysctl(net, "net/core", tbl);
 422         if (net->core.sysctl_hdr == NULL)
 423                 goto err_reg;
 424
 425         return 0;
 426
 427 err_reg:
 428         if (tbl != netns_core_table)
 429                 kfree(tbl);
 430 err_dup:
 431         return -ENOMEM;
 432 }
 433
 434 static __net_exit void sysctl_core_net_exit(struct net *net)
 435 {
 436         struct ctl_table *tbl;
 437
 438         tbl = net->core.sysctl_hdr->ctl_table_arg;
 439         unregister_net_sysctl_table(net->core.sysctl_hdr);
 440         BUG_ON(tbl == netns_core_table);
 441         kfree(tbl);
 442 }
 443
 444 static __net_initdata struct pernet_operations sysctl_core_ops = {
 445         .init = sysctl_core_net_init,
 446         .exit = sysctl_core_net_exit,
 447 };
 448
 449 static __init int sysctl_core_init(void)
 450 {
 451         register_net_sysctl(&init_net, "net/core", net_core_table);
 452         return register_pernet_subsys(&sysctl_core_ops);
 453 }
 454
 455 fs_initcall(sysctl_core_init);